基于引导滤波的图像序列光流计算技术研究

发布时间：2018-04-06 23:39

本文选题：光流算法　切入点：引导滤波　出处：《南昌航空大学》2017年硕士论文

【摘要】：光流是指物体或者三维场景在二维投影平面上像素运动的瞬时速度,其不仅携带了被观察物体的运动信息,还包含了物体的三维结构信息。因此图像序列光流计算技术研究是图像处理、模式识别与计算机视觉等领域的一个重要研究方向。21世纪以来,随着计算机软、硬件及显示技术的迅猛发展,图像序列光流计算技术已在社会生产及生活中的各个领域突显出了越来越重要的地位和作用,研究成果被广泛应用于气象分析、交通视频监控、无人自动驾驶系统、医学图像分割以及机器人视觉系统等领域。近年来,随着光流计算研究的不断深入,针对光流计算模型和优化方法的新理论与新技术不断涌现,图像序列光流计算技术在计算精度和计算效率等方面取得了突破性的进展。但是当图像序列中包含运动边缘模糊、非刚性运动、亮度突变与不连续等困难场景和运动形式时,光流计算的准确性和鲁棒性问题还有待进一步研究。本文主要是针对图像序列运动模糊等问题进行深入研究,通过设计基于引导滤波的图像序列光流计算优化模型,克服由复杂边缘结构、大位移和非刚性运动所引起的光流计算边缘模糊等问题,以改善光流计算的准确性和鲁棒性。本文的主要工作包括以下几点:1.对图像序列光流计算技术的研究背景与研究现状进行了详细的介绍,并对现阶段光流计算技术所面临的主要问题与挑战进行了详细的分析。2.重点对HS、CLG-TV等经典光流计算方法和Classic+NL、NNF等近年来涌现的代表性光流计算模型进行介绍,并分析各类计算方法的优缺点及适用范围。3.针对复杂边缘结构、大位移运动以及非刚性运动引起的边缘模糊等问题,提出图像金字塔分层细化框架下基于引导滤波的光流计算优化方法。在光流计算过程中,通过引导滤波的线性化作用,能够很好的保留图像边缘细节部分,克服了图像边缘模糊等问题,使得图像边缘达到增强和锐化的效果,提高了光流的计算精确性和鲁棒性。4.分别采用Middlebury、MPI Sintel和KITTI数据库提供的标准测试图像集对HS、CLG-TV、Classic+NL和NNF等典型变分光流方法进行引导滤波优化对比测试。实验结果表明,本文所基于引导滤波的光流计算优化模型能够有效地提高光流计算的准确性与鲁棒性,尤其是当图像序列中包含复杂边缘结构和大位移运动时,能够大幅提升光流计算的效果。
[Abstract]:Optical flow refers to the instantaneous velocity of pixels moving on the two-dimensional projection plane of an object or 3D scene. It not only carries the motion information of the object under observation, but also contains the three-dimensional structure information of the object.Therefore, the research of image sequence optical flow computing technology is an important research direction in the field of image processing, pattern recognition and computer vision. Since the 21st century, with the rapid development of computer software, hardware and display technology,Image sequence optical flow computing technology has been playing an increasingly important role in various fields of social production and life. The research results have been widely used in meteorological analysis, traffic video surveillance, autonomous driving systems.Medical image segmentation and robot vision system and other fields.In recent years, with the development of optical flow calculation, new theories and techniques for optical flow calculation models and optimization methods have been emerging, and the image sequence optical flow computing technology has made a breakthrough in computing accuracy and efficiency.However, the accuracy and robustness of optical flow computation need to be further studied when the image sequences contain difficult scenes and motion forms such as motion edge blur, non-rigid motion, brightness mutation and discontinuity.In this paper, the motion blur of image sequence is studied deeply, and the optimization model of image sequence optical flow based on guided filter is designed to overcome the complex edge structure.In order to improve the accuracy and robustness of optical flow calculation, large displacement and non-rigid motion caused by optical flow calculation edge blur and other problems.The main work of this paper includes the following points: 1.The research background and present situation of image sequence optical flow computing technology are introduced in detail, and the main problems and challenges faced by optical flow computing technology are analyzed in detail.The classical optical flow calculation methods such as HSN CLG-TV and the representative optical flow calculation models such as Classic NLNNF in recent years are introduced, and the advantages and disadvantages of various calculation methods and their application range are analyzed.Aiming at the problems of edge blur caused by complex edge structure, large displacement motion and non-rigid motion, an optical flow optimization method based on guided filter is proposed in the framework of image pyramid thinning.In the process of optical flow calculation, by guiding the linearization of filter, the details of image edge can be preserved well, and the problems such as edge blur can be overcome, so that the image edge can be enhanced and sharpened.The calculation accuracy and robustness of optical flow are improved.The typical variational optical flow methods such as HSN CLG-TV Classic NL and NNF are optimized by using standard test image sets provided by the Middlebury MPI Sintel and KITTI databases, respectively.The experimental results show that the optical flow calculation optimization model based on guided filter can effectively improve the accuracy and robustness of optical flow calculation, especially when the image sequence contains complex edge structure and large displacement motion.It can greatly improve the effect of optical flow calculation.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】